High-Performance Computing with the Weather Research and Forecasting System Model: A Case Study under Stable Conditions over Mexico Basin

Abstract

This study explores the performance of the Weather Research and Forecasting System Model (WRF v.4.0) for a winter case under stable meteorological conditions in the Mexico Basin. To evaluate the sensitivity to spatial resolution and parameterization configurations, a suite of different numerical experiments is designed to test five Planetary Boundary Layer (PBL) schemes coupled to a Surface Layer parameterization (SL) and a cloud microphysics (MP) parameterization to find an optimal configuration in terms of closeness to physical reality and computational efficiency. The WRF atmospheric dynamics core and its ancillary physics routines constitute a massively parallel FORTRAN code that runs on the Tlaloc cluster at the ICAyCC-UNAM with optimized MPICH software. Two model performance metrics are used: 1) Taylor statistics to measure the distance between simulations and observed meteorological fields (near-surface and upper-level temperature and winds), and 2) CPU execution time. Results show that the Mellor-Yamada-Janjic (M) scheme performs best near the surface at 2.0 km horizontal resolution. However, the Yonsei University (Y) PBL scheme outperforms the M scheme when looking at temperature vertical profiles at the exact horizontal resolution. Both PBL schemes show negligible CPU execution time differences.